1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
|
#include <Storages/MergeTree/MergeTreeSink.h>
#include <Storages/MergeTree/MergeTreeDataPartInMemory.h>
#include <Storages/StorageMergeTree.h>
#include <Interpreters/PartLog.h>
#include <DataTypes/ObjectUtils.h>
#include <Common/ProfileEventsScope.h>
namespace ProfileEvents
{
extern const Event DuplicatedInsertedBlocks;
}
namespace DB
{
struct MergeTreeSink::DelayedChunk
{
struct Partition
{
MergeTreeDataWriter::TemporaryPart temp_part;
UInt64 elapsed_ns;
String block_dedup_token;
ProfileEvents::Counters part_counters;
};
std::vector<Partition> partitions;
};
MergeTreeSink::~MergeTreeSink() = default;
MergeTreeSink::MergeTreeSink(
StorageMergeTree & storage_,
StorageMetadataPtr metadata_snapshot_,
size_t max_parts_per_block_,
ContextPtr context_)
: SinkToStorage(metadata_snapshot_->getSampleBlock())
, storage(storage_)
, metadata_snapshot(metadata_snapshot_)
, max_parts_per_block(max_parts_per_block_)
, context(context_)
, storage_snapshot(storage.getStorageSnapshotWithoutData(metadata_snapshot, context_))
{
}
void MergeTreeSink::onStart()
{
/// It's only allowed to throw "too many parts" before write,
/// because interrupting long-running INSERT query in the middle is not convenient for users.
storage.delayInsertOrThrowIfNeeded(nullptr, context, true);
}
void MergeTreeSink::onFinish()
{
finishDelayedChunk();
}
void MergeTreeSink::consume(Chunk chunk)
{
if (num_blocks_processed > 0)
storage.delayInsertOrThrowIfNeeded(nullptr, context, false);
auto block = getHeader().cloneWithColumns(chunk.detachColumns());
if (!storage_snapshot->object_columns.empty())
convertDynamicColumnsToTuples(block, storage_snapshot);
auto part_blocks = storage.writer.splitBlockIntoParts(block, max_parts_per_block, metadata_snapshot, context);
using DelayedPartitions = std::vector<MergeTreeSink::DelayedChunk::Partition>;
DelayedPartitions partitions;
const Settings & settings = context->getSettingsRef();
size_t streams = 0;
bool support_parallel_write = false;
for (auto & current_block : part_blocks)
{
ProfileEvents::Counters part_counters;
UInt64 elapsed_ns = 0;
MergeTreeDataWriter::TemporaryPart temp_part;
{
ProfileEventsScope scoped_attach(&part_counters);
Stopwatch watch;
temp_part = storage.writer.writeTempPart(current_block, metadata_snapshot, context);
elapsed_ns = watch.elapsed();
}
/// If optimize_on_insert setting is true, current_block could become empty after merge
/// and we didn't create part.
if (!temp_part.part)
continue;
if (!support_parallel_write && temp_part.part->getDataPartStorage().supportParallelWrite())
support_parallel_write = true;
String block_dedup_token;
if (storage.getDeduplicationLog())
{
const String & dedup_token = settings.insert_deduplication_token;
if (!dedup_token.empty())
{
/// multiple blocks can be inserted within the same insert query
/// an ordinal number is added to dedup token to generate a distinctive block id for each block
block_dedup_token = fmt::format("{}_{}", dedup_token, chunk_dedup_seqnum);
++chunk_dedup_seqnum;
}
}
size_t max_insert_delayed_streams_for_parallel_write = DEFAULT_DELAYED_STREAMS_FOR_PARALLEL_WRITE;
if (!support_parallel_write || settings.max_insert_delayed_streams_for_parallel_write.changed)
max_insert_delayed_streams_for_parallel_write = settings.max_insert_delayed_streams_for_parallel_write;
/// In case of too much columns/parts in block, flush explicitly.
streams += temp_part.streams.size();
if (streams > max_insert_delayed_streams_for_parallel_write)
{
finishDelayedChunk();
delayed_chunk = std::make_unique<MergeTreeSink::DelayedChunk>();
delayed_chunk->partitions = std::move(partitions);
finishDelayedChunk();
streams = 0;
support_parallel_write = false;
partitions = DelayedPartitions{};
}
partitions.emplace_back(MergeTreeSink::DelayedChunk::Partition
{
.temp_part = std::move(temp_part),
.elapsed_ns = elapsed_ns,
.block_dedup_token = std::move(block_dedup_token),
.part_counters = std::move(part_counters),
});
}
finishDelayedChunk();
delayed_chunk = std::make_unique<MergeTreeSink::DelayedChunk>();
delayed_chunk->partitions = std::move(partitions);
++num_blocks_processed;
}
void MergeTreeSink::finishDelayedChunk()
{
if (!delayed_chunk)
return;
for (auto & partition : delayed_chunk->partitions)
{
ProfileEventsScope scoped_attach(&partition.part_counters);
partition.temp_part.finalize();
auto & part = partition.temp_part.part;
bool added = false;
/// It's important to create it outside of lock scope because
/// otherwise it can lock parts in destructor and deadlock is possible.
MergeTreeData::Transaction transaction(storage, context->getCurrentTransaction().get());
{
auto lock = storage.lockParts();
storage.fillNewPartName(part, lock);
auto * deduplication_log = storage.getDeduplicationLog();
if (deduplication_log)
{
const String block_id = part->getZeroLevelPartBlockID(partition.block_dedup_token);
auto res = deduplication_log->addPart(block_id, part->info);
if (!res.second)
{
ProfileEvents::increment(ProfileEvents::DuplicatedInsertedBlocks);
LOG_INFO(storage.log, "Block with ID {} already exists as part {}; ignoring it", block_id, res.first.getPartNameForLogs());
continue;
}
}
added = storage.renameTempPartAndAdd(part, transaction, lock);
transaction.commit(&lock);
}
/// Part can be deduplicated, so increment counters and add to part log only if it's really added
if (added)
{
auto counters_snapshot = std::make_shared<ProfileEvents::Counters::Snapshot>(partition.part_counters.getPartiallyAtomicSnapshot());
PartLog::addNewPart(storage.getContext(), PartLog::PartLogEntry(part, partition.elapsed_ns, counters_snapshot));
storage.incrementInsertedPartsProfileEvent(part->getType());
/// Initiate async merge - it will be done if it's good time for merge and if there are space in 'background_pool'.
storage.background_operations_assignee.trigger();
}
}
delayed_chunk.reset();
}
}
|
